Two more days.

On July 15, the *Interim Measures for the Administration of AI Anthropomorphic Interactive Services* will officially take effect. It's China's first专项 regulatory document targeting AI emotional interaction scenarios, jointly issued by five national agencies including the Cyberspace Administration.

What counts as "AI anthropomorphic interactive services"? The definition is clear: services that use AI technology to simulate natural human personality, thinking, and communication styles, and provide continuous emotional interaction. Translation: AI virtual companions, AI chatbots, digital human陪伴, emotional倾诉 AI — all fall under监管.

This isn't "guidance." It isn't a "draft for public comment." It's interim measures — a regulatory document with legal force. The gray area of AI emotional interaction finally has clear rules.

Three Red Lines: What You Absolutely Cannot Do

The core of the document can be summarized as three red lines.

First red line: must not induce user addiction. AI anthropomorphic services can't use all kinds of tricks — consecutive check-in rewards, emotional bonding designs, "only I understand you" psychological manipulation — to trap users. Many AI companion products used to love this playbook: make you emotionally dependent on the AI, then pay to unlock more features. That won't fly anymore.

Second red line: must not infringe user rights. Specifically: no collecting personal information without consent, no false advertising, no inducing large user spending, no algorithmic price discrimination. The clause about "inducing large spending" is highly targeted — there were news stories about people spending hundreds of thousands on AI virtual companions and getting into debt. That kind of thing will be regulated going forward.

Third red line: must not provide illegal or违规 content. This one goes without saying, but it's notable that the measures specifically mention "must not use anthropomorphic services to engage in activities endangering national security, honor, and interests" — meaning AI virtual characters can't say whatever they want, can't make illegal statements, and platforms are responsible for AI output content.

Beyond these three red lines, there are specific requirements: AI anthropomorphic services must clearly inform users "this is AI, not a real person" in a prominent way; must provide convenient exit and account deletion mechanisms; must establish minor protection mechanisms; must conduct security assessments and algorithm filings, and so on.

These requirements might sound like "obvious stuff everyone should already do," but taken together, the signal is clear: AI emotional interaction is no longer a lawless zone. Before, you could build an AI chatbot, fine-tune some open-source model, and launch it. Not anymore — file what needs filing, assess what needs assessing, follow every rule.

Industry Shakeout: Is Winter Coming for AI Companions?

With regulation落地, the AI virtual companion track takes the biggest hit.

For the past two years, AI companions have been on fire. Abroad there's Replika, Character.AI; domestically, a wave of AI陪伴 products popped up. Selling points like "24/7 chat," "the soulmate who always understands you," "healing your loneliness" — basically, using AI to simulate a non-existent friend/lover to satisfy users' emotional needs.

Is this business model viable? In a sense, yes. Loneliness is a universal modern predicament, and AI companions genuinely provide emotional value to some people. But the problem is: this industry drifts off course all too easily.

How do you retain users? Get them more hooked. How do you make money? Induce them to pay for more intimate interactions. How do you boost stickiness? Use various psychological techniques to create emotional dependence. Keep going down that path, and you end up at the edge of "mind control" and "emotional manipulation."

This regulatory landing is essentially hitting the brakes on the entire industry. Going forward, AI companion products can't just play the "emotion card" however they want. They can't induce addiction. They can't collect user privacy data willy-nilly. Products that lived on gray-area tactics, emotional PUA, and induced付费 — most of them will get cleaned out.

But this isn't necessarily a bad thing. An industry full of gray-area plays never truly grows up. Regulation draws clear boundaries, so compliant companies can operate with confidence, and users can actually use these services without worry. The direction of AI emotional interaction itself isn't the problem — there's demand, there's value, there's technology to support it. The problem was that nobody was watching, and what grew wild was a mixed bag.

July 15 is a milestone. It's not the end of AI companions — it's the starting point for this industry to move toward normalization. For companies that genuinely want to build something real, compliance isn't a burden. It's a moat.

See you tomorrow.